Research On Intelligent Control Method Based On Recurrent Wavelet Neural Network For Wastewater Treatment Process

The wastewater treatment process（WWTP）is an important way to alleviate the water crisis,reduce the need for natural water and weaken water environment pollution.To achieve efficient and stable operation of the WWTP and to ensure the effluent quality meets the standards,it is critical to implement accurate control for WWTP.However,several complex physical and biochemical reactions are included in the WWTP.Therefore,there are many challenges in the process tracking control,as follows:1)The reaction mechanism of the WWTP is complex,and it is difficult to establish the accurate mechanism model.Thus,it is difficult to measure the key effluent quality parameters of WWTP,such as the ammonia nitrogen and biochemical oxygen demand.2)The external environment of the WWTP can only be passively accepted,and the influent flow fluctuates greatly.Moreover,there is a serious coupling between the control variables,and the system operation is in non-sationary.Therefore,it is a great challenge to realize the accurate control of WWTP.To solve the above problems,the intelligent control method based on the recurrent wavelet neural network is proposed in this thesis.Firstly,the mechanism and characteristics of biochemical reactions in WWTP are analyzed,and the soft computing model of effluent ammonia nitrogen in WWTP is constructed by extracting the corresponding characteristic variables.Secondly,using the historical data of effluent ammonia nitrogen,the prediction method is designed based on the phase space reconstruction to extract the features of the data and dynamic characteristics of ammonia nitrogen time series.Moreover,to deal with the dynamicity of WWTP,the univariate self-organizing control method is proposed to achieve the accurate control of dissolved oxygen concentration.Finally,the multivariate intelligent control strategy is designed and the corresponding wastewater treatment process control system is eatablished to ensure the efficient and stable operation of WWTP.The main research contents and innovation points of the thesis are as follows:（1）Soft computing model of effluent ammonia nitrogen in WWTPTo solve the problem of difficult real-time measurement of effluent ammonia nitrogen（NH₄-N）,the soft computing model based on self-organizing cascade neural network is proposed.Firstly,the characteristic variables of effluent NH₄-N are selected based on the principle component analysis to determine the input variables of the soft computing method.Secondly,based on the orthogonal least squares method and test errors,the neuron selection mechanism and the stopping criterion are designed to automatically adjust the structure of network.Then,the incremental learning algorithm is proposed to update the parameters of the model.Finally,the experimental results show that the proposed method can achieve real-time measurement of effluent NH₄-N and improve the accuracy of measurement.（2）Offline prediction method for effluent ammonia nitrogen based on SPRWNNTo deal with the nonlinearity of WWTP,the prediction of effluent NH₄-N is trasformed into a time series prediction problem.Then,the offline prediction method of effluent NH₄-N is designed based on data and the self-organizing pipelined recurrent wavelet neural network（SPRWNN）.Firstly,the self-organizing mechanism is designed based on the spiking strength to realize the automatic adjustment of hidden neurons in each module.Secondly,the module growing algorithm based on the prediction performance is proposed to solve the problem of network structure design.Then,the convergence of the network is proved theoretically.Finally,the experimental results show that the SPRWNN can not only automatically adjust the network structure,but also improves the prediction accuracy compared with the network with fixed structure.（3）Online prediction method for effluent NH₄-N based on phase space reconstructionTo solve the problem that the time-varying characteristics of WWTP lead to the degraded performance of effluent NH₄-N prediction model,the online prediction method based on the phase space reconstruction（PSR）and pipelined recurrent wavelet neural network（PRWNN）is proposed.Firstly,the chaotic characteristics of effluent NH₄-N time series are proved by the correlation dimension mechanism.Based on the chaotic characteristics,the phase space of effluent NH₄-N concentration is reconstructed by PSR technique.Secondly,the relationship model between inputs and output of reconstructed phase space is established by PRWNN.Then,the online gradient algorithm with adaptive learning rates is designed to update the parameters of model online.Finally,the results show that the proposed method can effectively improve the prediction accuracy of effluent NH₄-N concentration.（4）Dissolved oxygen concentration control method based on OG-PRWNNTo deal with the uncertainty of WWTP,the online-growing PRWNN（OG-PRWNN）controller is proposed to achieve accurate control of dissolved oxygen（DO）concentration.Firstly,the online growing mechanism is designed based on the control performance to automatically adjust the number of modules of the controller.Secondly,to meet the accuracy requirements of the control,the online algorithm combining adaptive learning rates is designed to train the controller.Then,the stability of the OG-PRWNN is analyzed by Lyapunov stability theorem when the controller structure is fixed and changed,respectively.Finally,simulation results show that the OG-PRWNN controller can ensure smooth and accurate tracking control performance.（5）Multi-variable decoupled control method for WWTPFor the coupling characteristics between control varaibles of WWTP,the self-organizing recurrent wavelet neural network controller incorporating information of input variables is designed to achieve multivariate accurate control of the WWTP.Firstly,to solve the coupling problem of DO and nitrate nitrogen（NO）,the joint input mehniasm is established based on the control errors.Secondly,the self-organizing mechanism based on the spiking strength of wavelet nodes is designed to automatically adjust the structure of controller.Then,the online learning algorithm combining adaptive learning rates is used to update the parameters of the controller.Finally,the stability of controller is proved by Lyapunov stability theorem.Experimental results demonstrate that the proposed control method can effectively improve the control accuracy of the WWTP.（6）Multi-variable self-organizing control system for WWTPTo address the problem that the WWTP is difficult to operate smoothly,the multi-variable control system based on self-organizing recurrent wavelet neural network is proposed for WWTP.Firstly,considering the computation load of control process,the multi-input multi-output recurrent wavelet neural network is used as the controller to control both DO concentration and NO concentration.Secondly,according to the spiking strength of wavelet basis,the self-organizing mechanism is designed to dynamically adjust the structure of controller.Then,to ensure its application in the actual process,the Lyapunov stability theorem is used to analyze the stability of the controller.Finally,the multi-variable self-organizing control system for WWTP is constructed based on the controller and the prediction model.The simulation experiments show that the control system can accurately control both DO concentration and NO concentration.